Generally, a voice recognition apparatus performs voice recognition with reference to words (a vocabulary) registered into a voice recognition dictionary. Although many words need to be registered into the voice recognition dictionary in order to improve the voice recognition rate, registration of many words into the one voice recognition dictionary results in longer time to perform a search for a word through the voice recognition dictionary when performing voice recognition, and therefore increase in the time required to carry out the voice recognition processing (that is, the voice recognition processing speed decreases). In contrast, although decrease in the number of words registered into the voice recognition dictionary reduces the time required to perform a search for a word through the voice recognition dictionary, the voice recognition rate decreases.
To solve this problem, there has been provided a voice recognition apparatus which has a plurality of voice recognition dictionaries, and which is so constructed as to reduce the time required to perform a search for a word through each of the voice recognition dictionaries by searching through each of the voice recognition dictionaries in turn and to improve the voice recognition rate.
For example, there has been provided a voice recognition apparatus which compares the features of a voice inputted by an unspecified speaker with those of a voice stored in a voice dictionary file which is selected from a plurality of voice dictionary files (voice recognition dictionaries), which, when trying to detect a word whose matching degree reaches a reference value, if there does not exist any word whose matching degree reaches a reference value, stores, as a word candidate, a word having a relatively-high matching degree, switches to another voice dictionary file, compares the features of the inputted voice with those of an inputted voice corresponding to the word candidate stored in the switched-to voice dictionary file, and then determines a matching degree between them, and which, when any word whose matching degree reaches the reference value cannot be found out of the voice dictionary file which was selected for the first time, repeatedly carries out comparison with the inputted voice while switching among the voice dictionary files, and continues the recognition process of finding out a correct answer (for example, refer to patent reference 1).
Furthermore, there has been provided a voice recognition apparatus which, in order to change the specification of a voice recognition dictionary and carry out recognition processing again after it has performed an erroneous recognition, stores the inputted voice in a voice storage unit, performs the recognition processing by using a voice recognition unit, and displays a recognition result, and which, when the user provides an instruction for switching among recognition dictionaries because the result is an erroneous recognition, makes a recognition dictionary control unit switch among the voice recognition dictionaries by using a recognition result control unit and then makes the voice recognition unit perform the recognition processing on the voice stored in the voice storage unit (for example, refer to patent reference 2).    [Patent reference 1] JP, 6-318097, A (pp. 2 to 3 and FIGS. 1 to 4)    [Patent reference 2] JP, 2000-200093, A gazette (pp. 3 to 5 and FIGS. 1 to 5)
Because the conventional voice recognition apparatus which is constructed as mentioned above switches among voice recognition dictionaries according to the features (a tone) of a voice inputted by an unspecified speaker and simply compares the features of the inputted voice with those of a voice stored in a voice recognition dictionary while switching among the voice recognition dictionaries until the degree of matching between the features of the inputted voice and those of a voice stored in a voice recognition dictionary reaches a reference value, the conventional voice recognition apparatus must store many words in each of the voice recognition dictionaries in order to improve the recognition rate of the inputted voice even if it has the plurality of voice recognition dictionaries. A problem with the conventional voice recognition apparatus is therefore that it is difficult to reduce the time required to make a search for a word through the voice recognition dictionaries.
Furthermore, when the displayed voice recognition result is an erroneous recognition, the conventional voice recognition apparatus switches among the voice recognition dictionaries according to the user's switching instruction and performs the voice recognition again, though the user has to view the displayed voice recognition result and input an instruction for switching among the voice recognition dictionaries when judging that the voice recognition result indicates erroneous recognition. Thus, the operation for acquiring a voice recognition result indicating correct recognition of the inputted voice is troublesome, and it takes much time for the user to acquire the correct voice recognition result. Particularly, the user of a voice recognition apparatus for use in a car navigation apparatus may be unable to concentrate on driving if he or she has to input an instruction for switching among voice recognition dictionaries while he or she drives.
The present invention is made in order to solve the above-mentioned problems, and it is therefore an object of the present invention to provide a voice recognition apparatus which is easy to operate and which can reduce the voice recognition processing time and can improve the recognition rate.